Guide Image

Guide Image is image-to-image generation in Sogni Pocket. Drop in a reference photo and the model uses it as a creative starting point — composition, palette, and mood carry through to the new render, with a Strength slider controlling how literal the result stays. It's the simplest way to make variations of an existing image, transfer a mood across styles, or anchor a generation without prescribing exact structure.

#What it does

Image-to-image, mobile-native. A reference image biases the entire generation, not just a masked region.
Whole-image guidance. The model picks up composition, color, lighting, and mood from the reference.
Distinct from ControlNet. Guide Image uses the raw pixels of the reference as a starting noise; ControlNet uses derived feature maps (edges, pose, depth, face). Reach for Guide Image for vibe and composition; reach for ControlNet for specific structural constraints.

#How to add a Guide Image

Open the generation screen in Pocket.
Tap the branch button to open the Guide Image picker.
Choose an image from your photo library or from your Sogni history.
Adjust the Strength slider (see below).
Type a prompt — what you want the new image to become — and tap Imagine.

The result will share composition and tone with the reference, reshaped by the prompt and selected model.

#Strength

The Strength slider controls how closely the result hugs the reference.

High strength (60–95%) — Result stays very close to the reference. Small reinterpretations, mostly variations.
Mid strength (40–60%) — Recognizable composition, looser color and texture. Good default for style transfer.
Low strength (10–40%) — The reference becomes a hint rather than a template; the prompt drives more of the result.

The number of denoising steps is automatically reduced as Strength drops, since the model has less work to do when it's preserving more of the input.

#Common uses

Style transfer. Keep a photograph's composition, render it as oil paint, ink, anime, or claymation.
Mood and palette matching. Use a reference for color and lighting; let the prompt redraw the subject.
Composition control without ControlNet. When you just want "something framed like this," not "this exact pose."
Iterating on a render. Send a finished Sogni image back in as a Guide Image to refine, restyle, or recolor.
Variations of a subject. Lower strength + the same prompt produces a series of related but distinct takes.

#Pairing with Paintover

Once you have a result you mostly like, switch to Paintover to fine-tune specific regions — emphasize details, hide unwanted elements, or tweak lighting in a single spot. The two are designed to compose. Tip: keep Paintover strength below 50% for the cleanest blends.

#Preventing degradation

When the Guide Image was generated by the same model with no additional guidance, reusing the same seed across passes can lead to visual degradation as the image is fed back into the same conditioning. Shuffle the seed once if you notice quality loss, then lock it again when you find a stable result.

#Tips

Start mid-strength. 40–60% is a forgiving range. Adjust up if the model is wandering, down if it's too literal.
Match aspect ratio when possible. A 1:1 reference into a 16:9 canvas forces awkward cropping in the conditioning step.
Use the gallery as a feedback loop. Pick a winner, send it back in as a new Guide Image at low strength, iterate.
Don't confuse it with ControlNet. If you need a specific pose or sketch, that's ControlNet. Guide Image is for "feel like this."
Shuffle the seed if quality drifts. Especially across repeated same-seed passes.